Practical guide to regulating a medical AI product

Medical AI products require certification before deployment in most jurisdictions. To date, no clear pathways for regulating medical AI exist. I present guidelines for development of such a regulatory package. This approach is predicated on the translation between a statistical risk perspective, typical of medical device regulators, and a deep understanding of machine learning methodologies. This work of translation will enable medical device regulators and machine learning experts to communicate more clearly, and thus lead to the development of standardised pathways for medical AI regulation.